Custom YOLOv12 Object Detection Model

I specialize in fine-tuning object detection models, including YOLOv4 (Darknet framework) and YOLOv12 (Ultralytics PyTorch framework), for robotics, UAVs, and autonomous perception systems. My work covers the complete pipeline: dataset preparation, hyperparameter tuning, model training, performance evaluation, and embedded deployment. Customized YOLO architectures for domain-specific datasets, adjusting hyperparameters (learning rate, batch size, input resolution, IoU thresholds, confidence scores) to maximize precision and recall. Well-versed with advanced fine-tuning options like augment, degrees, scale  etc. Deployed trained YOLO models into ROS-based robotic pipelines for real-time perception systems for object detection. Beyond YOLO, I have also trained DNN object detection models in Grounding Dino. 

I have also deployed large-scale deep neural networks on NVIDIA Jetson through a conversion pipeline from PyTorch .pt → ONNX → TensorRT, achieving efficient inference for real-world robotics applications.